atrium any% faq

Q: what the hell is this
A: atrium (T)WR attempts.

it's really bad, 3 second IL that is basically a string of frame perfect inputs aka some atari dragster bullshit.  the optimal run is 5 frame perfect inputs (broken down later) and current WR (3.183) is only 1 frame of leniency off a perfect run for humans - at the time, it was the best possible time for humans until i found a new optimization.  my current PB is 3.200, which was also the best possible time for humans with standard strats when set.

route breakdown, frame perfect inputs in parentheses: dash+jump first frame after countdown (1), downdash first possible frame (2), first frame dash cancel 2nd heavy (3), first frame downdash (4), final heavy (5).

the good news: everything in between those is buffers or relatively lenient jump spacing, and a lot of the frame perfect inputs are just muscle memory (consistent first frame movement + frame perfect heavy cancel downdashes are not uncommon for top players), it's pretty braindead and you just finish runs until it happens

the bad news: everything else.  the route above is an oversimplification and the best possible time you can get is a 3.200, aka my PB.  what the record does instead for the 3rd frame perfect input is a dash cancel before the attack hitstop begins, which is also frame perfect and lets you downdash 1 frame earlier.  the problem with this is it's frame perfect, there's absolutely no feedback on whether you pressed the button early or late (unless you check your inputs in a replay, which is time consuming and annoying), breaks the normal muscle memory you have for dash canceling heavies, and also breaks muscle memory for the downdash input afterwards (the first dash is now significantly earlier and you have to wait through hitstop + dash cooldown until you can downdash).  record does this and get everything significant on the first possible frame for a 3.183.

re: the optimization i found: turns out if you don't buffer a wallclimb for the topmost walljump, you save a frame due to being ever so slightly lower so you can do the final heavy earlier.  the good news is this adds no frame perfect inputs and is as easy to execute as simply not pressing up when doing the topmost walljump, the bad news is immediately after you want to do an upheavy so it's really easy to press up early and get the wallclimb and lose the timesave.  i probably won't reset over this mostly because i could still finish out a .183 and muscle memory for ending is important.

so if all goes well, everything frame perfect, 3.166 and untied WR.  3.183 most likely and would be 1 frame off perfect.



Q: why
A: few reasons -

1: my PB is corrupted and crashes your game if you watch it (unless you've watched another replay beforehand, in which case it just replays that).  pretty shitty because it's 2nd on the boards (first 3.200 ever achieved) and since it was a run i was pretty proud of at the time i still feel pretty fucked over by it and wanted to fix that
2: there's been some debate about the legitimacy of the WR, i'm leaning towards probably legit but the guy who got it interacted with the community very very little at the time and has pulled off a lot of other really dumb strats in questionably low timeframes.
3: kind of overdue for someone to tie it, 30 month old untied record
4: i kind of like this level, i played it a lot to warm up back in the day and it's just kind of mindless



Q: is 3.166 the best possible time for TAS?/any other possible timesaves for humans?
A: no/maybe.  TAS gets a 3.067 using a reverse ledgejump for the first downdash of the run, which is not human viable (tech has been known for 4 years now and a human has never executed it - tl;dr you need TAS level ledgecancels and need to switch from left -> right in a frame in the middle of a 2 frame ledgecancel (holding left+right in dustforce results in a neutral input)).  this saves a bunch of time and would probably kill the level if someone managed it lol.

TAS has the optimization of downdashing at the same frame as the topmost walljump - this makes your walljump not have jump startup frames but means you're jumping a lot lower from the wall.  this saves a frame (2?) but means you can no longer buffer the topmost heavy out of walljump startup frames, adds an extra frame perfect input, and is really, really fucking hard.  not sure if humans can do this but it's basically the only avenue of improvement.  TAS also loses some times to cycles so it's saving slightly less to humans than it should be.

lastly, with double tap dash you can actually save 2 frames at the start with some tech that shortens jump squat frames, but it adds an extra frame perfect input and makes the heavy afterwards really, really, really hard to the point im really hesitant to say it's human viable.



Q: are you playing more dustforce?
A: probably not no.  this mech sucks and can't dashjump consistently at no fault of my own so playing more than half of the ILs in the game is a crapshoot.  i'm mildly interested in playing again but not happening any time soon, especially considering this keyboard is great for celeste imo (a game that actually has sensible input buffering so needing to press actions on the same exact frame is a lot less meaningful) so i have no reason to drop 80+ bucks and play the mech lottery again.  probably upgrading my computer after the end of the year and i think my issue might be a power supply problem so maybe that'll magically fix my problem.